Video Extraction and Tracking of Moving Targets with a Camera

Cameras are nowadays widely used in maritime surveillance applications to detect small non-cooperative boats. This paper presents a simple and efficient video extractor based upon the background subtraction approach to obtain a binary mask of the moving pixels which are then segmented into targets by means of a clustering step. For the first step, we use a robust Kalman filter to obtain the binary mask with an Expectation / Maximization step to update the model's parameters. Then, an adaptive bandwidth mean shift approach is used to obtain the final clusters corresponding to the detected objects, comprising real targets and false alarms. Finally, a multitarget tracking algorithm fed by the bounding box centers corresponding to the detected objects is used to track the different targets in the image plane and to filter out the false alarms. Real results in a maritime surveillance context are presented.